The frequency that gets an answer, and the quarter cycle nobody mentions

A child on a swing is pushed once per swing and goes higher every time — the same pendulum as ever, with one term added to its equation. A wine glass held near a loudspeaker at the right pitch begins to sing and can be shattered. A radio picks one station out of the hundreds arriving at its aerial simultaneously. All three are the same phenomenon, and all three are usually explained with a sentence that is wrong in an interesting way.

The response, as a function of what is asked for #

An oscillator that is driven and damped settles, after any transient has died away, into a motion at exactly the frequency it is being driven at — not at its own. That is the first thing worth stating, because it is often assumed the opposite way round.

What depends on the driving frequency is not the rate but the size of the response, and the phase relationship between the driving and the responding. Writing $r$ for the ratio of driving frequency to natural frequency and $\zeta$ for the damping ratio, the amplitude relative to the deflection a steady force of the same size would produce is

$$\frac{1}{\sqrt{(1 - r^2)^2 + (2\zeta r)^2}},$$

and every feature of the figure is one of those two terms winning.

Far below the natural frequency, $r \approx 0$ and the response is 1: the oscillator simply follows the force, as a spring would follow a hand moving slowly. Far above, the $r^2$ dominates and the response falls off as $1/r^2$: the oscillator cannot keep up, and at high enough frequency it barely moves at all. Between the two, the $(1 - r^2)$ term passes through zero, and if the damping is small there is nothing left in the denominator to stop the response becoming very large.

That is resonance: a term in a denominator going to zero, with only the damping preventing a division by nothing.

The peak is not where the name suggests #

The maximum is at

$$r = \sqrt{1 - 2\zeta^2},$$

which is below the natural frequency, and by an amount the figure prints for each curve. At $\zeta = 0.05$ the peak is at 0.9975 of the natural frequency; at $\zeta = 0.4$ it is at 0.825; and beyond $\zeta = 1/\sqrt{2} \approx 0.707$ there is no peak at all — the response falls monotonically from its zero-frequency value and the system does not resonate in any useful sense.

That last threshold is a design specification rather than a curiosity. A car suspension is damped at around $\zeta = 0.3$, which keeps a peak but a modest one; damping it below about 0.2 gives a car that wallows at one particular speed over a rippled road, and damping it above 0.707 gives one that transmits every bump directly. Instruments meant to follow a signal faithfully — an accelerometer, a galvanometer, a moving-coil meter — are damped deliberately near 0.7, because that is the value at which the response stays flattest over the widest band.

The distinction between the natural frequency, the damped natural frequency and the resonant frequency is genuinely three different numbers. They agree to a fraction of a per cent in anything lightly damped, which is why the distinction is usually skipped, and they diverge exactly where the damping is large enough to matter.

The quarter cycle that does the work #

Here is the part that the swing explanation gets wrong, and correcting it turns resonance from a slogan into a mechanism.

The usual account says the pushes “add up” because they are in step with the motion. If that were right, the force would be in phase with the displacement — and at resonance it is not. It is a quarter cycle behind the displacement, which puts it exactly in phase with the velocity.

That is the mechanism, and it is an energy argument. The work done by a force is force times velocity, added over the cycle. If the force is in phase with the displacement, it pushes forward for half the cycle and backward for the other half, and the work over a full cycle is zero. If the force is in phase with the velocity, it pushes forward whenever the motion is forward, so the work is positive at every instant of the cycle, and the energy delivered per cycle is as large as it can possibly be. Work is force along the motion, and this is the arrangement that maximises it.

So resonance is the condition in which the driver never once opposes the motion. The amplitude then grows until the energy delivered per cycle equals the energy lost per cycle to damping, and that balance is what fixes the steady-state height of the peak.

The quarter-cycle lag also explains something a swing makes obvious and the formula does not. Pushing a swing works best at the bottom of the arc, where the child is moving fastest — not at the top, where the displacement is greatest. Anyone who has done it knows this without knowing why. The why is that the useful push is the one aligned with velocity.

The sharpness, and the number that measures it #

Lightly damped systems have tall narrow peaks and heavily damped ones have short broad peaks, and the two properties are not independent — they are the same property.

The quality factor $Q = 1/2\zeta$ measures both. The peak height is approximately $Q$, and the fractional width of the peak, measured between the points where the power has fallen by half, is approximately $1/Q$. A system with $Q = 100$ responds a hundred times as strongly at resonance and does so over a band one per cent wide.

The range across physical systems is enormous. A car suspension has $Q \approx 1.7$. A plucked guitar string, maybe 1,000. A wine glass, around 1,000 — which is why it can be shattered by a voice, and also why the voice has to be within about a tenth of a per cent of the right pitch, which is the real difficulty of the trick. A quartz crystal reaches $10^4$ to $10^6$, and that is the entire reason quartz replaced the pendulum in timekeeping: a resonance that narrow is a frequency reference. The mirror suspensions in a gravitational-wave detector reach $10^8$, deliberately, so that their thermal vibration is concentrated into a narrow band that can be avoided rather than spread across the measurement.

What a sharp resonance costs #

$Q$ is not free, and the charge is paid in time.

A high-$Q$ oscillator stores energy well and therefore loses it slowly — that is the same statement — so it takes a long time to build up and a long time to die away. The amplitude reaches most of its steady-state value only after roughly $Q/\pi$ cycles of driving. For a quartz crystal at $Q = 10^5$ that is around 30,000 cycles, which at 32,768 Hz is about a second: a quartz watch does not start ticking accurately the instant power is applied.

That trade is the same one that stops a short note from having a definite pitch, and it is not a coincidence — it is the same theorem. A narrow response in frequency is a slow response in time, always, with the product bounded below. There is no such thing as a filter that is both sharply selective and quick to react, in any technology, at any price.

The consequence is visible whenever both properties are wanted at once. A radio’s intermediate-frequency filter has $Q \approx 100$ at 10.7 MHz, giving a bandwidth of about 100 kHz — wide enough to pass the music and narrow enough to reject the adjacent station — and its ring-up time is a few microseconds, which is invisible. Push $Q$ up to sharpen the selectivity and the filter starts to smear the signal it is passing, because the ring-up time approaches the timescale of the modulation. Every filter design is a position taken on this trade, and the trade is the same one that governs how sharply a source’s spectrum can be measured.

The second cost is fragility. A sharp resonance is sharp in both directions: it must be hit accurately to be excited, and it must be hit accurately to be avoided. A structure with a lightly damped mode at 4.2 Hz — a frequency the boundaries decide rather than the designer — is enormously vulnerable to anything periodic at 4.2 Hz and indifferent to 4.5 Hz. Engineering practice is therefore usually to add damping rather than to move the frequency, since damping lowers the peak everywhere and moving the frequency merely relocates the problem.

Where the model stops #

The curve above is the steady-state response of a linear oscillator with one degree of freedom, and each of those qualifiers fails somewhere familiar.

Steady state. The formula describes the motion after the transient has died, and says nothing about the approach. A system driven at resonance from rest has an amplitude that grows roughly linearly at first — and an undamped system driven exactly at resonance grows without limit and never reaches a steady state at all, which is the case the formula reports as a division by zero.

Linear. The restoring force was assumed proportional to displacement. A real pendulum’s period depends on its amplitude, so as the response grows the natural frequency shifts, and the peak leans over — driving up in frequency then gives a different curve from driving down, with a sudden jump between them. That hysteresis is characteristic of nonlinear resonance and there is no sign of it in any linear treatment.

One degree of freedom. A real structure has many modes, each with its own frequency and damping, and the response is the sum. A building, a bridge or an instrument body has a forest of peaks, and the modes are the standing waves the boundaries permit.

And the driving must be at a frequency of its own. The most consequential failure in the list is that not every catastrophic vibration is resonance. The Tacoma Narrows bridge is presented as the standard example of resonance destroying a structure, and it was not one: the wind was steady, at no particular frequency, and the deck’s motion generated the forces that fed it — a self-excited oscillation called flutter, in which the energy input depends on the motion rather than on any external period. The distinction matters because the remedies differ. Adding damping helps against resonance and may not help against flutter, which has to be cured by changing the shape so the aerodynamic feedback loses its sign. Aircraft are certified against flutter specifically, and by a different analysis.

A second common misattribution is worth the same treatment. A microwave oven does not work by resonating water molecules. Water’s rotational resonances are in the hundreds of gigahertz; ovens run at 2.45 GHz, far below any of them, on the broad shoulder of the absorption rather than on a peak. If the oven did work at resonance the outside of the food would absorb everything and the middle would stay frozen, which is the opposite of the design intent. The frequency was chosen for penetration depth and for the licensing of the band.

Two things resonance is not #

Two failures of the concept are worth separating out, because both are commonly taught as examples of it and neither is one.

A system driven off resonance still moves. The curve does not fall to zero anywhere; it falls off as $1/r^2$ above and settles at 1 below. So the presence of motion is not evidence of resonance, and the absence of a matching frequency is not a guarantee of safety. A structure with a mode at 4 Hz driven at 8 Hz responds about a quarter as strongly as it would under a steady force — small, and not nothing, and cumulative if the loading is repeated for long enough to matter to fatigue.

Growth without limit needs zero damping, which does not exist. The image of an oscillator driven at resonance until it destroys itself belongs to the undamped case, and every real system reaches a steady amplitude of about $Q$ times the static deflection. That number is the honest measure of the danger: a structure with $Q = 30$ driven at resonance experiences thirty times the deflection a static load of the same size would produce. Enough to break most things, and finite, and calculable in advance.

The practical consequence of both is that resonance is a design quantity rather than a design event. What matters is not whether a resonance exists — it always does, at many frequencies — but how large $Q$ is at each of them, and whether anything in the environment supplies energy near those frequencies for long enough for the response to build.

The same peak, everywhere #

The reason a curve about a mass on a spring deserves a whole essay is that almost nothing in it is about masses or springs.

The identical mathematics governs a mass on a spring, a pendulum, an electrical circuit with inductance and capacitance, an acoustic cavity, a molecule’s vibrational mode, an atom’s electronic transition, a nuclear spin in a magnetic field, and an orbit perturbed periodically by another body. Each has its own constants; the shape of the curve is the same, and $Q$ means the same thing in every one.

Some of the consequences of that universality are large. Magnetic resonance imaging tips nuclear spins by driving them at their resonant frequency in a field, and reads the frequency to locate the signal — which requires a $Q$ high enough to distinguish tissues and a field gradient to convert position into frequency. Spectroscopy of every kind is resonance: shine a range of frequencies at something and record where it absorbs, and the absorption lines are its resonances, each with a width that reports how quickly it loses energy — which is why the width of a line also reports the temperature of the gas, once the two contributions are separated. The width of a spectral line is a $Q$ measurement, and reading it gives lifetimes that no clock could measure directly.

And the orbital case produces structure at astronomical scale. Gaps in the asteroid belt sit at radii whose orbital period is a simple ratio of Jupiter’s, because a repeated tug at the same phase of every orbit accumulates in exactly the way the swing does. The Kirkwood gaps are resonance drawn across the solar system, and they are empty for the same reason the peak in the first figure is tall.

The ladder from here #

Later rungs: the transient solution, and how the response builds. Resonance in the electrical case, where $Q$ has an elementary expression in the component values. Coupled oscillators, where two resonances at the same frequency split into two at different ones — the avoided crossing, which is one of the most reused structures in physics. Parametric resonance, driven by varying a property of the oscillator rather than pushing it, which is how a swing is actually pumped by someone standing on it. Nonlinear resonance and the leaning peak. Flutter and self-excited oscillation treated properly. Damping mechanisms, and why internal friction, radiation and viscous drag give different frequency dependences. And the resonances of continuous systems, where the discrete list of modes replaces the single natural frequency assumed throughout this essay.